[SOLVED] GSEA - Strange Results interpretation - Error in DESEQ2 ?
0
0
Entering edit mode
3 months ago
Evan Seffar ▴ 140

Hi, I'm currently doing GSEA with my list of differentially expressed genes obtained with DeSeq2. I'm new in this kind of analysis and I'm currently encountering strange results.

To give you some background, I compared two basic conditions : treated vs untreated. I would like to know which genes are differentially expressed in the treatment condition. This treatment in the first condition induce apoptosis and it's known for that. I'm trying to find the genes which has been up / down regulated and the pathway impacted which are responsible of apoptosis.

The fact is, when I'm doing GSEA following the tutorial of ClusterProfiler, the pathways associated with apoptosis are down regulated. It seems to be inverse to what happens biologically.

Do you have any ideas to check if I didn't done something wrong ? Maybe I didn't compare my conditions in the right way with DeSeq2 and the over expressed genes are down expressed genes in fact ?

In DeSeq2 I made this contrast :

res <- results(dds, contrast = "group", "treated","untreated")


By doing this does DeSeq2 gives me over expressed genes in treated condition or untreated condition ? The Log2(FC) are calculated as treatment over control or the other way around ?

Thank you

gsea deseq2 clusterprofiler • 295 views
1
Entering edit mode

Are you sure that that is your original code? shouldn't it be:

res <- results(dds, contrast = c("group", "treated","untreated"))

That command should give you log2(treated/untreated). Nonetheless, because any package will be different, your annotations in the design may be change, etc. I would suggest to always look at your data to check: e.g. plot the counts from the top significant genes across your groups, to see if what you will label as up- or down-regulated is indeed so.

Other possibilities: either there is a mistake in other steps (group annotation, specifying the model design, or in the use ClusterProfiler, etc.) which is impossible to know by your information, or the results are indeed biologically unexpected.

0
Entering edit mode

Thank you very much for taking time to answer. This is the right code yes, I forgot a ")" pasting the code.

I check the counts and It seems that when my Log2FC in less than -1.5, the counts in my treated samples are below than the untreated condition. So the Log2FC calculated is correct and corresponds to what I was waiting for.

So the fact is, it seems there aren't error in Deseq2. Maybe in GSEA so ? I followed this tutorial step by step : http://yulab-smu.top/clusterProfiler-book/

I verified my code but it is the exact same than in the tutorial. I will try with another package to be sure I'm not doing any errors. Could you advice a good website where I could put my differentially expressed genes list to see if I obtain the same pathways than in clusterprofiler ?

1
Entering edit mode

Hi Evan!

To perform GSEA analysis I suggest you some important points:

• Use another R package such as fgsea which is more stricter and curtomizable than ClusterProfiler (at least for me)

• According to the information released from the GSEA software, they suggest to run the analysis using the non-pre-ranked method (i.e. using normalized counts)

• Use the entire list of your genes, i.e. those remaining after your abundance filter to perform GSEA. According to this review you should avoid to use results from differential expression analysis to perform GSEA.

Best regards!

0
Entering edit mode

Hi ! Thank you very much for your help ! I will try this package :)